Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Capturing Momentum: Tennis Match Analysis Using Machine Learning and Time Series Theory

Jingdi Lei, Tianqi Kang, Yuluan Cao, Shiwei Ren

TL;DR

This study investigates whether momentum in tennis can be quantified and leveraged for predictive insights. It combines Hidden Markov Models to infer latent momentum states with Exponential Moving Average to compute a momentum signal, and then evaluates its utility using XGBoost and LightGBM, complemented by SHAP-based feature interpretation. The results indicate that explicitly modeling momentum improves predictive performance and reveals actionable indicators such as net points won, break points, and elapsed time. The approach demonstrates generalization across major tournaments and offers practical guidance for match preparation and targeted training.

Abstract

This paper represents an analysis on the momentum of tennis match. And due to Generalization performance of it, it can be helpful in constructing a system to predict the result of sports game and analyze the performance of player based on the Technical statistics. We First use hidden markov models to predict the momentum which is defined as the performance of players. Then we use Xgboost to prove the significance of momentum. Finally we use LightGBM to evaluate the performance of our model and use SHAP feature importance ranking and weight analysis to find the key points that affect the performance of players.

Capturing Momentum: Tennis Match Analysis Using Machine Learning and Time Series Theory

TL;DR

This study investigates whether momentum in tennis can be quantified and leveraged for predictive insights. It combines Hidden Markov Models to infer latent momentum states with Exponential Moving Average to compute a momentum signal, and then evaluates its utility using XGBoost and LightGBM, complemented by SHAP-based feature interpretation. The results indicate that explicitly modeling momentum improves predictive performance and reveals actionable indicators such as net points won, break points, and elapsed time. The approach demonstrates generalization across major tournaments and offers practical guidance for match preparation and targeted training.

Abstract

This paper represents an analysis on the momentum of tennis match. And due to Generalization performance of it, it can be helpful in constructing a system to predict the result of sports game and analyze the performance of player based on the Technical statistics. We First use hidden markov models to predict the momentum which is defined as the performance of players. Then we use Xgboost to prove the significance of momentum. Finally we use LightGBM to evaluate the performance of our model and use SHAP feature importance ranking and weight analysis to find the key points that affect the performance of players.
Paper Structure (28 sections, 7 equations, 17 figures, 1 table)

This paper contains 28 sections, 7 equations, 17 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Problem Statement
  3. Related Work
  4. Our Work
  5. Preparation for the Models
  6. Data Preprocessing
  7. Momentum Quantification
  8. Model Selection
  9. Hidden Markov model
  10. HMM Construction
  11. Exponential Moving Average
  12. The Match Flow
  13. Performance Metrics
  14. Assessing the role of Momentum
  15. Xgboost Regression Model
  16. ...and 13 more sections

Figures (17)

  • Figure 1: 2023 Wimbledon Gentlemen's final scores
  • Figure 2: Our work and models in the whole process. A is data cleaning . B presents the method to find the momentum. C shows the method to prove the significance of momentum. D find the importance feature in momentum change. E evaluate the performance of our model
  • Figure 4: The first four rows of preprocessed data
  • Figure 5: Bayes and HMM
  • Figure 6: Schematic diagram of the hidden Markov model
  • ...and 12 more figures